huyuanfeng2018 commented on code in PR #904: URL: https://github.com/apache/flink-kubernetes-operator/pull/904#discussion_r1822219343
########## flink-autoscaler/src/main/java/org/apache/flink/autoscaler/ParallelismAdjuster.java: ########## @@ -0,0 +1,148 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.flink.autoscaler; + +import org.apache.flink.autoscaler.config.AutoScalerOptions; +import org.apache.flink.autoscaler.event.AutoScalerEventHandler; +import org.apache.flink.autoscaler.topology.ShipStrategy; +import org.apache.flink.configuration.DescribedEnum; +import org.apache.flink.configuration.description.InlineElement; +import org.apache.flink.runtime.jobgraph.JobVertexID; + +import java.util.Collection; + +import static org.apache.flink.autoscaler.JobVertexScaler.SCALE_LIMITED_MESSAGE_FORMAT; +import static org.apache.flink.autoscaler.JobVertexScaler.SCALING_LIMITED; +import static org.apache.flink.autoscaler.ParallelismAdjuster.KeyGroupOrPartitionsAdjustMode.MAXIMIZE_UTILISATION; +import static org.apache.flink.autoscaler.config.AutoScalerOptions.SCALING_EVENT_INTERVAL; +import static org.apache.flink.autoscaler.topology.ShipStrategy.HASH; +import static org.apache.flink.configuration.description.TextElement.text; + +/** + * Component responsible adjusts the parallelism of a vertex. + * + * <p>When input vertex {@link ShipStrategy} is {@link ShipStrategy#HASH} or knows the number of + * current partitions of vertex. We hope to adjust the parallelism of the current vertex according + * to the number of key groups or partitions to achieve the goal of evenly distributing data among + * subtasks or maximizing utilization. + */ +public class ParallelismAdjuster { + + public static <KEY, Context extends JobAutoScalerContext<KEY>> int adjust( + JobVertexID vertex, + Context context, + AutoScalerEventHandler<KEY, Context> eventHandler, + int maxParallelism, + int numSourcePartitions, + int newParallelism, + int upperBound, + int parallelismLowerLimit, + Collection<ShipStrategy> inputShipStrategies) { + var adjustByMaxParallelismOrPartitions = + numSourcePartitions > 0 || inputShipStrategies.contains(HASH); + if (!adjustByMaxParallelismOrPartitions) { + return newParallelism; + } + var numKeyGroupsOrPartitions = + numSourcePartitions <= 0 ? maxParallelism : numSourcePartitions; + + KeyGroupOrPartitionsAdjustMode mode = + context.getConfiguration() + .get(AutoScalerOptions.SCALING_KEY_GROUP_PARTITIONS_ADJUST_MODE); + + var upperBoundForAlignment = Math.min(numKeyGroupsOrPartitions, upperBound); + + // When the shuffle type of vertex inputs contains keyBy or vertex is a source, + // we try to adjust the parallelism such that it divides + // the numKeyGroupsOrPartitions without a remainder => data is evenly spread across subtasks + for (int p = newParallelism; p <= upperBoundForAlignment; p++) { + if (numKeyGroupsOrPartitions % p == 0 + || + // When Mode is MAXIMIZE_UTILISATION , Try to find the smallest parallelism + // that can satisfy the current consumption rate. + (mode == MAXIMIZE_UTILISATION + && numKeyGroupsOrPartitions / p + < numKeyGroupsOrPartitions / newParallelism)) { + return p; + } + } + + // When adjusting the parallelism after rounding up cannot + // find the right degree of parallelism to meet requirements, + // Try to find the smallest parallelism that can satisfy the current consumption rate. + int p = + calculateMinimumParallelism( + numKeyGroupsOrPartitions, newParallelism, parallelismLowerLimit); + var message = + String.format( + SCALE_LIMITED_MESSAGE_FORMAT, + vertex, + newParallelism, + p, + numKeyGroupsOrPartitions, + upperBound, + parallelismLowerLimit); + eventHandler.handleEvent( + context, + AutoScalerEventHandler.Type.Warning, + SCALING_LIMITED, + message, + SCALING_LIMITED + vertex + newParallelism, + context.getConfiguration().get(SCALING_EVENT_INTERVAL)); + return p; + } + + private static int calculateMinimumParallelism( + int numKeyGroupsOrPartitions, int newParallelism, int parallelismLowerLimit) { + int p = newParallelism; + for (; p > 0; p--) { + if (numKeyGroupsOrPartitions / p > numKeyGroupsOrPartitions / newParallelism) { + if (numKeyGroupsOrPartitions % p != 0) { + p++; + } + break; + } + } + p = Math.max(p, parallelismLowerLimit); + return p; + } + + /** The mode of the key group or parallelism adjustment. */ + public enum KeyGroupOrPartitionsAdjustMode implements DescribedEnum { + DEFAULT( + "This mode ensures that the parallelism adjustment attempts to evenly distribute data across subtasks" + + ". It is particularly effective for source vertices that are aware of partition counts or vertices after " + + "'keyBy' operation. The goal is to have the number of key groups or partitions be divisible by the set parallelism, ensuring even data distribution and reducing data skew."), Review Comment: okļ¼great advice. ########## flink-autoscaler/src/main/java/org/apache/flink/autoscaler/ParallelismAdjuster.java: ########## @@ -0,0 +1,148 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.flink.autoscaler; + +import org.apache.flink.autoscaler.config.AutoScalerOptions; +import org.apache.flink.autoscaler.event.AutoScalerEventHandler; +import org.apache.flink.autoscaler.topology.ShipStrategy; +import org.apache.flink.configuration.DescribedEnum; +import org.apache.flink.configuration.description.InlineElement; +import org.apache.flink.runtime.jobgraph.JobVertexID; + +import java.util.Collection; + +import static org.apache.flink.autoscaler.JobVertexScaler.SCALE_LIMITED_MESSAGE_FORMAT; +import static org.apache.flink.autoscaler.JobVertexScaler.SCALING_LIMITED; +import static org.apache.flink.autoscaler.ParallelismAdjuster.KeyGroupOrPartitionsAdjustMode.MAXIMIZE_UTILISATION; +import static org.apache.flink.autoscaler.config.AutoScalerOptions.SCALING_EVENT_INTERVAL; +import static org.apache.flink.autoscaler.topology.ShipStrategy.HASH; +import static org.apache.flink.configuration.description.TextElement.text; + +/** + * Component responsible adjusts the parallelism of a vertex. + * + * <p>When input vertex {@link ShipStrategy} is {@link ShipStrategy#HASH} or knows the number of + * current partitions of vertex. We hope to adjust the parallelism of the current vertex according + * to the number of key groups or partitions to achieve the goal of evenly distributing data among + * subtasks or maximizing utilization. + */ +public class ParallelismAdjuster { + + public static <KEY, Context extends JobAutoScalerContext<KEY>> int adjust( + JobVertexID vertex, + Context context, + AutoScalerEventHandler<KEY, Context> eventHandler, + int maxParallelism, + int numSourcePartitions, + int newParallelism, + int upperBound, + int parallelismLowerLimit, + Collection<ShipStrategy> inputShipStrategies) { + var adjustByMaxParallelismOrPartitions = + numSourcePartitions > 0 || inputShipStrategies.contains(HASH); + if (!adjustByMaxParallelismOrPartitions) { + return newParallelism; + } + var numKeyGroupsOrPartitions = + numSourcePartitions <= 0 ? maxParallelism : numSourcePartitions; + + KeyGroupOrPartitionsAdjustMode mode = + context.getConfiguration() + .get(AutoScalerOptions.SCALING_KEY_GROUP_PARTITIONS_ADJUST_MODE); + + var upperBoundForAlignment = Math.min(numKeyGroupsOrPartitions, upperBound); + + // When the shuffle type of vertex inputs contains keyBy or vertex is a source, + // we try to adjust the parallelism such that it divides + // the numKeyGroupsOrPartitions without a remainder => data is evenly spread across subtasks + for (int p = newParallelism; p <= upperBoundForAlignment; p++) { + if (numKeyGroupsOrPartitions % p == 0 + || + // When Mode is MAXIMIZE_UTILISATION , Try to find the smallest parallelism + // that can satisfy the current consumption rate. + (mode == MAXIMIZE_UTILISATION + && numKeyGroupsOrPartitions / p + < numKeyGroupsOrPartitions / newParallelism)) { + return p; + } + } + + // When adjusting the parallelism after rounding up cannot + // find the right degree of parallelism to meet requirements, + // Try to find the smallest parallelism that can satisfy the current consumption rate. + int p = + calculateMinimumParallelism( + numKeyGroupsOrPartitions, newParallelism, parallelismLowerLimit); + var message = + String.format( + SCALE_LIMITED_MESSAGE_FORMAT, + vertex, + newParallelism, + p, + numKeyGroupsOrPartitions, + upperBound, + parallelismLowerLimit); + eventHandler.handleEvent( + context, + AutoScalerEventHandler.Type.Warning, + SCALING_LIMITED, + message, + SCALING_LIMITED + vertex + newParallelism, + context.getConfiguration().get(SCALING_EVENT_INTERVAL)); + return p; + } + + private static int calculateMinimumParallelism( + int numKeyGroupsOrPartitions, int newParallelism, int parallelismLowerLimit) { + int p = newParallelism; + for (; p > 0; p--) { + if (numKeyGroupsOrPartitions / p > numKeyGroupsOrPartitions / newParallelism) { + if (numKeyGroupsOrPartitions % p != 0) { + p++; + } + break; + } + } + p = Math.max(p, parallelismLowerLimit); + return p; + } + + /** The mode of the key group or parallelism adjustment. */ + public enum KeyGroupOrPartitionsAdjustMode implements DescribedEnum { + DEFAULT( + "This mode ensures that the parallelism adjustment attempts to evenly distribute data across subtasks" + + ". It is particularly effective for source vertices that are aware of partition counts or vertices after " + + "'keyBy' operation. The goal is to have the number of key groups or partitions be divisible by the set parallelism, ensuring even data distribution and reducing data skew."), + + MAXIMIZE_UTILISATION( + "This model is to maximize resource utilization. In this mode, an attempt is made to set" + + " the minimum degree of parallelism that meets the current consumption rate requirements. Unlike the default mode, it is not enforced that the number of key groups or partitions is divisible by the degree of parallelism."), Review Comment: ok. ########## flink-autoscaler/src/main/java/org/apache/flink/autoscaler/ParallelismAdjuster.java: ########## @@ -0,0 +1,148 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.flink.autoscaler; + +import org.apache.flink.autoscaler.config.AutoScalerOptions; +import org.apache.flink.autoscaler.event.AutoScalerEventHandler; +import org.apache.flink.autoscaler.topology.ShipStrategy; +import org.apache.flink.configuration.DescribedEnum; +import org.apache.flink.configuration.description.InlineElement; +import org.apache.flink.runtime.jobgraph.JobVertexID; + +import java.util.Collection; + +import static org.apache.flink.autoscaler.JobVertexScaler.SCALE_LIMITED_MESSAGE_FORMAT; +import static org.apache.flink.autoscaler.JobVertexScaler.SCALING_LIMITED; +import static org.apache.flink.autoscaler.ParallelismAdjuster.KeyGroupOrPartitionsAdjustMode.MAXIMIZE_UTILISATION; +import static org.apache.flink.autoscaler.config.AutoScalerOptions.SCALING_EVENT_INTERVAL; +import static org.apache.flink.autoscaler.topology.ShipStrategy.HASH; +import static org.apache.flink.configuration.description.TextElement.text; + +/** + * Component responsible adjusts the parallelism of a vertex. + * + * <p>When input vertex {@link ShipStrategy} is {@link ShipStrategy#HASH} or knows the number of + * current partitions of vertex. We hope to adjust the parallelism of the current vertex according + * to the number of key groups or partitions to achieve the goal of evenly distributing data among + * subtasks or maximizing utilization. + */ +public class ParallelismAdjuster { + + public static <KEY, Context extends JobAutoScalerContext<KEY>> int adjust( + JobVertexID vertex, + Context context, + AutoScalerEventHandler<KEY, Context> eventHandler, + int maxParallelism, + int numSourcePartitions, + int newParallelism, + int upperBound, + int parallelismLowerLimit, + Collection<ShipStrategy> inputShipStrategies) { + var adjustByMaxParallelismOrPartitions = + numSourcePartitions > 0 || inputShipStrategies.contains(HASH); + if (!adjustByMaxParallelismOrPartitions) { + return newParallelism; + } + var numKeyGroupsOrPartitions = + numSourcePartitions <= 0 ? maxParallelism : numSourcePartitions; + + KeyGroupOrPartitionsAdjustMode mode = + context.getConfiguration() + .get(AutoScalerOptions.SCALING_KEY_GROUP_PARTITIONS_ADJUST_MODE); + + var upperBoundForAlignment = Math.min(numKeyGroupsOrPartitions, upperBound); + + // When the shuffle type of vertex inputs contains keyBy or vertex is a source, + // we try to adjust the parallelism such that it divides + // the numKeyGroupsOrPartitions without a remainder => data is evenly spread across subtasks + for (int p = newParallelism; p <= upperBoundForAlignment; p++) { + if (numKeyGroupsOrPartitions % p == 0 + || + // When Mode is MAXIMIZE_UTILISATION , Try to find the smallest parallelism + // that can satisfy the current consumption rate. + (mode == MAXIMIZE_UTILISATION + && numKeyGroupsOrPartitions / p + < numKeyGroupsOrPartitions / newParallelism)) { Review Comment: Sorry for replying so late. > I understand the idea behind this but I wonder whether it has the desired effect. The autoscaling algorithm isn't aware of any state/partition imbalance, it assumes linear scale through state/partition balance can be achieved. A slight adjustment to the parallelism won't drastically improve the situation. When the partition or keygroups is evenly distributed and there are relatively many divisors such as (60, 120, 720), we can indeed easily achieve linear scaling through state/partition balancing, which is the most ideal situation. If the number of partitions is unreasonable, such as 35, hope adjust the parallelism to linearly increase the consumption rate, Although it is not possible to consume evenly, it is meaningful. Imagine a situation: when running with 7 parallelism degrees, the busy value of the operator is large. No matter what the `scaleFactor` , it can be expected that the autoscaler will expand the parallelism degree to 35, which may cause the operator to be very idle, but we know that the linear assumption may not be completely linear when changing 7->35, and it may not be possible to obtain a 5 (35/7) times increase in processing speed (this is related to many factors), it may be in the next cycle Trigger scale down again, and then the cycle repeats In addition, due to the limitations of the **`scale-down.max-factor`** parameter, we may never be able to reduce the degree of parallelism If use the MAXIMIZE_UTILISATION mode, can significantly improve this phenomenon. > It looks like this change could help in situations where the number of partition / key groups do not have many divisors, but its also kind of hard to reason about. Any suggestions? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
